基于融合特征和Voting集成学习的膜蛋白类型预测

Prediction of Membrane Protein Types Based on Fusion Feature Information and Voting Ensemble Learning

摘要: 膜蛋白是细胞功能的主要承担者，其功能与其类型密切相关。膜蛋白类型的鉴定是生物信息学中的一项重要课题。已有的膜蛋白分类模型主要从膜蛋白序列信息中提取特征，本文提出了一种基于蛋白质二级结构信息的蛋白质特征提取方法，并将其融入现有的两种序列特征。通过对比实验结果显示，在融入了蛋白质二级结构特征后，几种不同机器学习分类算法下的膜蛋白预测精度均有提升，说明了该融合蛋白质二级结构特征方法的有效性。最后，基于Voting集成学习框架，结合三种机器学习算法构建膜蛋白分类模型。结果表明，该模型的预测效果优于现有的几种机器学习模型。

Abstract: Studies have shown that membrane proteins are the main bearers of cellular functions and their functions are closely related to their types. Therefore, the identification of membrane protein types is an important topic in bioinformatics. The existing classification models for membrane proteins mainly extract features from the sequence information of membrane proteins. In this paper, a protein feature extraction method was proposed based on protein secondary structure information, which was integrated into two existing sequence features. By comparing the experimental results, the prediction accuracy of membrane proteins under several different machine learning classification algorithms was improved after integrating protein secondary structure features, which illustrated the effectiveness of this fusion protein secondary structure feature method. Finally, a membrane protein classification model was constructed based on the voting ensemble learning frame-work in combination with three machine learning algorithms. The results show that the prediction performance of this model is better than other machine learning models.